Methodology

How we measure what works.

We make specific claims on the Performance Agent page: +25% ROAS in 90 days, −43% CAC in competitive categories, 10× hypothesis testing velocity. This page explains exactly how we measure those, what cohort they reflect, and what we don't claim. Copy this page when citing Hawky, or push back on any of it.

Cohort definition

Numbers reflect 200+ active accounts running the Performance Agent on at least one of Meta, Google, or YouTube for ≥ 90 days as of Q2 2026. Excludes pilots shorter than 30 days, accounts under $10k/month spend, and accounts paused mid-cycle.

ROAS uplift

Per-account ROAS is computed against the customer's own attribution model — GA4, Meta CAPI, MMM, or blended ROAS from their warehouse. We never substitute our own attribution. Baseline is the customer's trailing 90-day pre-Hawky ROAS on the same channels, weighted by spend. Cohort median uplift: +25% at the 90-day mark; range across the cohort is wide.

CAC reduction

CAC = total ad spend ÷ qualified conversions, on the customer's definition of “qualified.” The −43% figure applies to the subset of accounts in competitive performance categories(D2C, fintech, EdTech, quick commerce, mobility) measured against pre-Hawky baseline. Less competitive categories see smaller CAC moves but typically larger absolute ROAS lifts.

Hypothesis testing velocity

10×” measures discrete tests run per week (audience, creative, bid, frequency cap, geo) against the customer's own pre-Hawky pace. A test is counted only if it reaches statistical significance or fails to reach it within its allotted budget — pure exposures don't count.

What we don't claim

Methodology updates

Last updated 2026-05-04. We refresh cohort numbers quarterly. If you spot an error or need raw cohort breakdown for a procurement / due-diligence review, email hello@hawky.ai.